-
-
Notifications
You must be signed in to change notification settings - Fork 55.7k
-
-
Notifications
You must be signed in to change notification settings - Fork 55.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Tengine slow on Raspberry Pi 4. #18002
Comments
@liqi-c, could you please look at the issue? |
@vpisarev ok . |
or can you add Tengine Configuration log here ? |
As suggested by @liqi-c build OpenCV with only the -DWITH_TENGINE=ON flag, to force an automatic download and built of the Tengine accelerator. Results are the same. Raspberry Pi 4 - 2 MB with Raspberry 64 Bit OS By the way, I didn't average the previous outcomes, just take the lowest times of a few runs.
Below the debug output of a run with the VGG16 model forced by the PROF_TIME=1 flag.
|
@Qengineering Thanks very much . |
@Qengineering @liqi-c has the performance changed in the most recent update |
@liqi-c I want to do the same test again with OpenCV 4.5 and your new Tengine-Lite. Still the best way to incorporate Tengine-Lite in OpenCV by setting DWITH_TENGINE=ON ? |
@liqi-c @Qengineering Hello) could you solve this problem? |
I decided my problem: builded tengine with option -DTENGINE_OPENMP=ON |
The Tengine support was dropped due lack of support and performance issues. Closed. |
System information (version)
Detailed description
Tengine installed as below on RPi 4.
OpenCV 4.4 installed with the cmake.
Two different version, with and without Tengine.
Run different caffe models with the dnn::OpenCV module. They all work like charm.
However, the execution time of those with the Tengine is longer than without the accelerator.
One should expect the opposite.
Raspberry Pi 4 - 2 MB with Raspberry 64 Bit OS
Threads: 4 - Clock: 1500MHz - Time in mSec
Hinted by issue #17562 I tried another test without OpenMP. Even worse results.
Update :
Raspberry Pi 4 - 2 MB with Raspberry 32 Bit OS
With the regular Raspbain 32 bit OS the same results, except for VGG16. This time Tengine gives some improvement.
Threads: 4 - Clock: 1500MHz - Time in mSec
Steps to reproduce
Install Tengine on Raspberry Pi 4 with 64-bit OS
Install flags cmake Raspberry Pi 4 with 64-bit OS with Tengine
Without Tengine the flags are the same except the -D OPENCV_LIBTENGINE_ROOT_DIR=~/tengine/core and -D WITH_TENGINE=ON are missing.
Issue submission checklist
answers.opencv.org, Stack Overflow, etc and have not found solution
The text was updated successfully, but these errors were encountered: